To properly utilize the nutraceutical properties of green tea, therefore, we need to clarify the relationship between cultivar and bioactivity. For nutraceutical evaluation, it is important to elucidate which cultivars have bioactivity, and which compounds contribute directly or indirectly to this bioactivity. In this study, we applied metabolic profiling techniques to evaluate the bioactivity of 43 representative cultivars of Japanese green tea. The aim of our research was to evaluate the relationship between the metabolome and bioactivity of diverse tea cultivars. To test bioactivity we investigated the ability of leaf extracts to inhibit thrombininduced MRLC in human umbilical vein endothelial cells , as a potential hallmark of MK-0683 vascular endothelial dysfunction. In addition, analyses of metabolic data from all tea extracts clearly discriminated green tea cultivars according to their bioactivity. Using regression analysis, we constructed a model to predict the bioactivity of tea cultivars on the basis of their metabolic data. These approaches comprise a useful strategy both for evaluation of bioactivity of green tea cultivars and for identification of bioactive factors. For all LC-MS Selumetinib datasets, data were processed using the free software XCMS to extract and align peaks. Total tea extracts , tea extracts from three cultivars , and two types of treated tea extracts were evaluated separately by multivariate statistical analysis. Generally, this analysis is used to clarify similarities and differences among samples on the basis of multivariate data. A multivariate approach can decrease the complexity of huge MS datasets, and can reveal relationships among samples or datasets. These relationships are usually displayed as scatter plots. Since hundreds of variables are obtained in MS analyses, the relationships among samples must be theoretically interpreted on hundreds of dimensional axes , but these relationships cannot be displayed simply. To visualize the features of samples, multivariate statistical analysis can extract features of samples by dimensional reduction. That is, hundreds of original variables are decreased to two or three synthetic variables, which are orthogonal with each other. This maximizes the statistical variance of samples, while leaving the original feature of samples largely unaffected. The synthetic variables consist of hundreds of original variables.